OpenAI launches Sora: how the new app will impact children's safety.

New parental controls from OpenAI

According to Vox: It is quite difficult to determine how many children have been harmed by AI-based chatbots, yet their presence is evident. For nearly three years, ChatGPT has been available to all age groups without any restrictions. But on Monday, OpenAI unveiled a new set of parental controls, some of which are aimed at preventing teenage suicides, as in the case of 16-year-old Adam Rayne from California, who died after talking with ChatGPT about how to do it. The next day, the company launched a social network with an app called Sora, which resembles TikTok but features 'hyper-realistic' videos created by AI.

Not coincidentally, OpenAI announced these parental controls alongside ambitious plans to compete with Instagram and YouTube. On one hand, the company released a new app to engage users with AI-based content, while on the other, it gave parents more control options. The new settings primarily relate to ChatGPT, although parents can also restrict what their children see in Sora.

Challenges of the innovations

However, the new controls for ChatGPT raise questions. Parents can now link their children's accounts to their own and impose restrictions on sensitive content. If at any point OpenAI's tools detect a serious safety risk, a moderator will review the situation and, if necessary, notify the parents. However, parents cannot read the transcripts of their children's conversations with ChatGPT, and teenagers can disconnect their accounts from parents at any time (OpenAI assures that parents will be notified of this).

Currently, it is not entirely clear how all of this will work in practice, but some experts believe that OpenAI is not doing enough to protect children. 'Even adults have problems with self-regulation when chatbots offer a fun and clingy friend available for conversation every hour,' says Robbie Torney, senior director of AI programs at Common Sense Media.

'Even adults struggle to self-regulate when AI chatbots offer a fun, sycophantic friend available for conversation every hour of the day.'

According to Torney, many people have become addicted to AI without even realizing it. 'They started using AI to help with homework or work, and slowly began to use it for other purposes.'

Risks for teenagers

This raises serious concerns, as more than 70% of teenagers use chatbots for communication. 'These are indeed serious and well-documented dangers,' according to a recent survey by Common Sense Media. That is why AI partner applications, such as Character.ai, already have certain default restrictions for young users.

Here lies an even broader problem: parental controls place the responsibility for protecting children on parents, not on the tech companies themselves. Parents typically need to dive into settings and change them. Moreover, parents remain responsible for how their children use these products.

The true goal of parental controls

It is worth mentioning that OpenAI introduced these controls and the new app at a time when an important AI safety bill was on the table of California Governor Gavin Newsom, which he signed on the same day the parental controls were announced. 'The true aim of these parental tools, whether it involves ChatGPT or Instagram, is not to protect children,' states Josh Golin, executive director of the organization Fairplay.

'It aims to say that self-regulation is fine. Leave us alone, please!'

'There are no parental controls that will guarantee safety 100%.'

This situation creates a difficult choice for parents. They must know that their child is using ChatGPT and make decisions about what settings will be sufficient to ensure safety but not too strict so that the child does not create a fake account.

At this point, OpenAI is trying to find a balance between user safety and the desire to engage them with its applications in the future. Time will tell how effective these innovations will be.

This version of the news was also published in the User Friendly newsletter.